lookup: evaluation tools, use corpus/previous gens

Author: JohannesGaessler

.gitignore

@@ -58,6 +58,9 @@ models-mnt
 /llava-cli
 /lookahead
 /lookup
+/lookup-create
+/lookup-merge
+/lookup-stats
 /main
 /metal
 /passkey

Makefile

@@ -676,14 +676,17 @@ json-schema-to-grammar.o: common/json-schema-to-grammar.cpp common/json-schema-t
 train.o: common/train.cpp common/train.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
+ngram-cache.o: common/ngram-cache.cpp common/ngram-cache.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+
 libllama.so: llama.o ggml.o $(OBJS)
 	$(CXX) $(CXXFLAGS) -shared -fPIC -o $@ $^ $(LDFLAGS)
 
 libllama.a: llama.o ggml.o $(OBJS) $(COMMON_DEPS)
 	ar rcs libllama.a llama.o ggml.o $(OBJS) $(COMMON_DEPS)
 
 clean:
-	rm -vrf *.o tests/*.o *.so *.a *.dll benchmark-matmult common/build-info.cpp *.dot $(COV_TARGETS) $(BUILD_TARGETS) $(TEST_TARGETS)
+	rm -vrf *.o tests/*.o *.so *.a *.dll benchmark-matmult lookup-create lookup-merge lookup-stats common/build-info.cpp *.dot $(COV_TARGETS) $(BUILD_TARGETS) $(TEST_TARGETS)
 	find examples pocs -type f -name "*.o" -delete
 
 #
@@ -813,9 +816,15 @@ lookahead: examples/lookahead/lookahead.cpp ggml.o llama.o $(COMMON_DEPS) $(OBJS
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
-lookup: examples/lookup/lookup.cpp ggml.o llama.o $(COMMON_DEPS) $(OBJS)
+lookup: examples/lookup/lookup.cpp ggml.o llama.o ngram-cache.o $(COMMON_DEPS) $(OBJS)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
+	$(CXX) $(CXXFLAGS) -c examples/lookup/lookup-create.cpp -o $(call GET_OBJ_FILE, examples/lookup/lookup-create.cpp)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, examples/lookup/lookup-create.cpp) -o lookup-create $(LDFLAGS)
+	$(CXX) $(CXXFLAGS) -c examples/lookup/lookup-merge.cpp -o $(call GET_OBJ_FILE, examples/lookup/lookup-merge.cpp)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, examples/lookup/lookup-merge.cpp) -o lookup-merge $(LDFLAGS)
+	$(CXX) $(CXXFLAGS) -c examples/lookup/lookup-stats.cpp -o $(call GET_OBJ_FILE, examples/lookup/lookup-stats.cpp)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, examples/lookup/lookup-stats.cpp) -o lookup-stats $(LDFLAGS)
 
 passkey: examples/passkey/passkey.cpp ggml.o llama.o $(COMMON_DEPS) $(OBJS)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)

common/CMakeLists.txt

@@ -65,6 +65,8 @@ add_library(${TARGET} STATIC
     json.hpp
     train.h
     train.cpp
+    ngram-cache.h
+    ngram-cache.cpp
     )
 
 if (BUILD_SHARED_LIBS)

common/common.cpp

@@ -963,6 +963,22 @@ static bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg,
         }
         return true;
     }
+    if (arg == "-lcs" || arg == "--lookup-cache-static") {
+        if (++i >= argc) {
+            invalid_param = true;
+            return true;
+        }
+        params.lookup_cache_static = argv[i];
+        return true;
+    }
+    if (arg == "-lcd" || arg == "--lookup-cache-dynamic") {
+        if (++i >= argc) {
+            invalid_param = true;
+            return true;
+        }
+        params.lookup_cache_dynamic = argv[i];
+        return true;
+    }
     if (arg == "--save-all-logits" || arg == "--kl-divergence-base") {
         if (++i >= argc) {
             invalid_param = true;
@@ -1436,6 +1452,10 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
     printf("                        Hugging Face model file (default: unused)\n");
     printf("  -ld LOGDIR, --logdir LOGDIR\n");
     printf("                        path under which to save YAML logs (no logging if unset)\n");
+    printf("  -lcs FNAME, --lookup-cache-static FNAME\n");
+    printf("                        path to static lookup cache to use for lookup decoding (not updated by generation)\n");
+    printf("  -lcd FNAME, --lookup-cache-dynamic FNAME\n");
+    printf("                        path to dynamic lookup cache to use for lookup decoding (updated by generation)\n");
     printf("  --override-kv KEY=TYPE:VALUE\n");
     printf("                        advanced option to override model metadata by key. may be specified multiple times.\n");
     printf("                        types: int, float, bool. example: --override-kv tokenizer.ggml.add_bos_token=bool:false\n");

common/common.h

@@ -88,20 +88,22 @@ struct gpt_params {
     // // sampling parameters
     struct llama_sampling_params sparams;
 
-    std::string model             = "models/7B/ggml-model-f16.gguf"; // model path
-    std::string model_draft       = "";  // draft model for speculative decoding
-    std::string model_alias       = "unknown"; // model alias
-    std::string model_url         = "";  // model url to download
-    std::string hf_repo           = "";  // HF repo
-    std::string hf_file           = "";  // HF file
-    std::string prompt            = "";
-    std::string prompt_file       = "";  // store the external prompt file name
-    std::string path_prompt_cache = "";  // path to file for saving/loading prompt eval state
-    std::string input_prefix      = "";  // string to prefix user inputs with
-    std::string input_suffix      = "";  // string to suffix user inputs with
+    std::string model                = "models/7B/ggml-model-f16.gguf"; // model path
+    std::string model_draft          = "";  // draft model for speculative decoding
+    std::string model_alias          = "unknown"; // model alias
+    std::string model_url            = "";  // model url to download
+    std::string hf_repo              = "";  // HF repo
+    std::string hf_file              = "";  // HF file
+    std::string prompt               = "";
+    std::string prompt_file          = "";  // store the external prompt file name
+    std::string path_prompt_cache    = "";  // path to file for saving/loading prompt eval state
+    std::string input_prefix         = "";  // string to prefix user inputs with
+    std::string input_suffix         = "";  // string to suffix user inputs with
     std::vector<std::string> antiprompt; // string upon seeing which more user input is prompted
-    std::string logdir            = "";  // directory in which to save YAML log files
-    std::string logits_file       = "";  // file for saving *all* logits
+    std::string logdir               = "";  // directory in which to save YAML log files
+    std::string lookup_cache_static  = ""; // path of static ngram cache file for lookup decoding
+    std::string lookup_cache_dynamic = ""; // path of dynamic ngram cache file for lookup decoding
+    std::string logits_file          = "";  // file for saving *all* logits
 
     std::vector<llama_model_kv_override> kv_overrides;